Context Engineering

Long-running agents strain context, reward design, and GPU scheduling at once. Compaction and value models can help, but each introduces another optimization trade-off.

Vercel MCP now serves the 2026-07-28 and 2025 protocols from one endpoint, letting newer clients adopt stateless requests and revised authorization without breaking older setups.

ReToken adds one learned retrieval embedding to select relevant visual tokens from a cached context, improving long-image and video benchmarks while fitting training and inference on one H100.

AISPA turns system-prompt review into an eight-dimension audit. Its survey suggests builders should test prompts for user protection and conflicting instructions, not merely check that safeguards exist.

Add agents beside deterministic fraud systems, not in place of them: project cross-service events into a semantic layer, send only uncertain cases to specialist agents, then emit the verdict back into the saga.

mcp-handler 2.0 adds the stateless 2026 MCP protocol and SDK v2 while keeping 2025 Streamable HTTP clients working, but drops HTTP+SSE and raises upgrade requirements.

Shared agents need a different harness: action-boundary security, selective group memory, per-user privacy, and routing that decides both who receives information and when the agent should speak.

FactSet treats skills as versioned product features and the harness as their runtime. Routing descriptions, model-specific evals, access controls, and governance matter as libraries grow.

Varick treats enterprise agents as process-reengineering systems: capture how work really happens, encode that context, then automate only the steps whose risk permits it.

Ramp pairs strict request scoping with agents across the delivery pipeline. Its intake agent reportedly saves about 20% of scoping time, while humans retain final judgment.

UniMem routes novel experience into retrieval memory and consolidates recurring patterns into expandable parameters, offering a design for agents that learn across unlabelled task streams.

Kubernetes security patches improved sharply when the model received live service dependencies, showing that scanner findings alone omit context needed to avoid breaking workloads.

Decagon splits deployment between configuring each customer’s agent and turning repeated enterprise requests into product features. The scarce skill is resisting brittle one-offs.

Kepler frames forward deployment as product discovery: observe real work, ship the smallest useful fix, then turn repeated pain and customer vocabulary into durable product leverage.

Graph-shaped context can expose relationships and document structure that vector search or Text-to-SQL misses. Treat it as an additional retrieval surface, then benchmark it against your own data.

LLM synthesis breaks simple source pointers when facts merge, change, or outlive one input. Model provenance inside the context graph so agents can filter trust, debug derivation, and apply deletion policies.

Atlan’s agent experiments argue for shared, versioned context instead of per-agent memory: a portable layer for business facts, skills, norms, retrieval, and feedback across changing harnesses.

E3 makes agents estimate task scope, try the minimum viable path, and expand only after verification fails. In a controlled edit benchmark, it preserved task completion while sharply reducing work.

Stable aggregate accuracy can hide individual answers flipping when irrelevant context is added. Agent evaluations should compare outputs per task and probe realistic context noise, not only average scores.

An open-source memory hub turns agent conversations, workflows, docs, and code into governed assets that can be reused across sessions and roles, reducing repeated project setup.

book-to-skill compiles books and document sets into on-demand agent skills, reducing repeated context loading while preserving chapter-level references and reusable decision rules.

A RAG study finds that retrieved ideology carries into answers and varies with sampling temperature, so source audits and decoding tests should be evaluated together.

RLMs treat a large repository as external data that an agent inspects with code, returning bounded evidence to the main context instead of loading or summarizing everything upfront.

Graphify gives coding agents a queryable project graph with provenance-tagged relationships, reducing repeated repository scans while keeping inferred links visibly distinct from extracted facts.

CompactionRL uses RL to teach agents to compact their own context mid-task, lifting GLM-4.5-Air 7 points to 66.8% on SWE-bench Verified; the recipe is now in GLM-5.2's training pipeline.

ReContext is a training-free harness that replays query-relevant evidence from long inputs before answering, taking the best average rank across 8 long-context benchmarks up to 128K on Qwen3-4B/8B and Llama3-8B.

Marketing Skills gives coding agents shared product context and task-specific workflows for CRO, copy, SEO, analytics, pricing, and launch work, reducing repeated setup across growth tasks.

A maintained archive of extracted system prompts from 100+ AI products — Claude Fable 5, Claude Code, ChatGPT 5.5, Gemini, Cursor — useful for studying how vendors actually instruct their agents.